#behavior-tree #tree-node #node-tree #ai #gamedev

bin+lib bhv

A crate providing means to create and use Behavior Trees

7 unstable releases (3 breaking)

0.4.0 Mar 3, 2024
0.3.0 Feb 24, 2024
0.2.3 Oct 25, 2023
0.1.2 Oct 23, 2023

#134 in Robotics

23 downloads per month

MIT license

56KB
960 lines

bhv

Bhv is a library for working with behavior trees.

But what are behavior trees?

A behavior tree is basically a tree where each node defines some action/function unique to the node type. Each node returns a "status" after they are executed, which determines whether the node has fully finished executing (and whether it executed successfully or not) or it still needs to be ran later. Parent nodes provide composition (think of a node that runs all of its children until one of them returns a "finished executing but not successful") while leaf nodes act as "atomic actions", ie. actions that cannot be split further (think of a simple condition check or a message being printed to the terminal). To share state between nodes, a "context" type is passed to each node being ran. Parent nodes usually adapt this context type by their children, while the leaf nodes are the ones that usually define a concrete type as context.

There is a set of nodes that are commonly used and thus provided by the library, specifically:

Adaptor nodes - leaf nodes that adapt functions into a node of the tree, such as cond that executes successfully when a condition is successful or action that simply executes a function and returns successfully.

Decorator nodes - nodes with a single child that manipulate the child's status or run it several times depending on some condition, such as inv that changes the execution status from success to failure and vice versa, or repeat(n) that runs a node until it is completed n times.

Composite nodes - nodes with at least one child that run them according to some condition, such as sequence that runs its child nodes as long as they execute successfully or selector that runs its child nodes until one of them executes successfully.

As it can be noticed, composition based on a node's status provides logic similar to control flow in programming languages. Furthermore, transition between states becomes as simple as providing a node with a condition and the state's nodes to be executed if the condition is successful (ie. just a sequence of the condition and then the other nodes). This makes behavior trees an alternative to state machines, where the states are not aware of each other as state management logic happens inside of parent nodes instead of child nodes.

For a more in-depth explanation and more examples, this GameAIPro chapter is a great resource.

Installation

Simple as running

cargo add bhv
# or for the `events` feature
cargo add bhv --features events

on the directory of the project where you want to use the library.

Features

The library provides several nodes that are commonly present in behavior trees. Specifically, the nodes provided are split into categories as following:

  • adaptor nodes, that adapt a function into a behavior tree node.
  • decorator nodes, that alter the behavior and result of a node.
  • composite nodes, that several nodes at the same time.

For help with specific nodes, refer to the documentation of the crate.

Showcase

// Guess the number game implemented using behavior trees.
use std::{
    io::{self, Write},
    time::{SystemTime, UNIX_EPOCH},
};

use bhv::*;

fn main() {
    game()
}

// The data that will be used by the nodes
#[derive(Default)]
struct Context {
    guess: u32,
    answer: u32,
}

// the nodes
struct RandomizeAnswer(u32, u32);
struct ReadInput;

macro_rules! print_msg {
    ($($arg:tt)*) => {
        action(|_ctx| print!($($arg)*))
    };
}

fn game() {
    let tree = seq! {
        RandomizeAnswer(0, 101),
        seq! {
            print_msg!("Enter a number from 0 to 100\n"),
            ReadInput,
            sel! {
                seq! {
                    cond(|ctx: &Context| ctx.guess < ctx.answer),
                    print_msg!("Your guess is smaller than the actual number\n").fail(), // The game is not over yet
                },
                seq! {
                    cond(|ctx: &Context| ctx.guess == ctx.answer),
                    print_msg!("Your guess is correct!\n"),
                },
                seq! {
                    cond(|ctx: &Context| ctx.guess > ctx.answer),
                    print_msg!("Your guess is greater than the actual number\n").fail(), // The game is not over yet
                }
            },
        }.repeat_until_pass(),
    };

    tree.execute(&mut Context::default());
}

impl Bhv for RandomizeAnswer {
    type Context = Context;

    fn update(&mut self, ctx: &mut Self::Context) -> Status {
        let time = SystemTime::now()
            .duration_since(UNIX_EPOCH)
            .unwrap_or_default()
            .as_secs();
        ctx.answer = (time % ((self.1 - self.0) as u64)) as u32 + self.0;

        Status::Success
    }
}

impl Bhv for ReadInput {
    type Context = Context;

    fn update(&mut self, ctx: &mut Self::Context) -> Status {
        io::stdout().flush().unwrap_or_default();

        let mut buff = String::new();

        io::stdin()
            .read_line(&mut buff)
            .map(|_| match buff.trim().parse() {
                Ok(v) => {
                    ctx.guess = v;
                    Status::Success
                }
                Err(e) => {
                    println!("Error reading from stdin :{}\t buffer: '{}'", e, buff);
                    Status::Failure
                }
            })
            .unwrap_or_else(|_| Status::Running)
    }
}

Features

Events

Version 0.3 introduces a new model for behavior trees, available under the feature flag events. This experimental feature allows nodes to run only when relevant event(s) occur. When this feature will be deemed complete, it will replace the actual default implementation, so it is recommended to migrate existing codebases to use this feature.

Compared to the "default" implementation, nodes on events introduce a new function called should_react_to(&self, kind: EventKind) -> bool that indicates whether an event of type kind is relevant to this node or not. By default, all events are considered relevant unless the behavior is overwritten. Additionally, the update function is now called react(&mut self, event: &dyn Event, ctx: &mut Self::Context), where the Event trait is used to make sure the values passed are meant to be used as event data. This model allows nodes to adapt to a more asynchronous event-based model of execution.

To reflect these changes, the execute function now accepts a new parameter that can be converted into an iterator that yields &dyn Event and be used as a stream of events for the node. This function is part of BhvExt instead of Bhv, meaning it can only have the default implementation and also that nodes can now be ?Sized.

As for differences in nodes provided, .repeat_until(cond) is NOT present when this feature is enabled. To achieve similar behavior with bhv.repeat_until(cond), please use seq! { bhv.pass(), cond }.repeat_until_pass(). Alternatively, you can use the new .wait_for::<EventType>() decorator that runs a node only if a certain event is triggered. To use this decorator the event should implement EventType. Think of EventType implementors as simply Event types that are not enum types.

To implement Event for your own types, proceed as following:

  • For non-enum types, simply implement EventType as
use bhv::*;

struct MyType; // my special event type

impl EventType for MyType {}

// Event is automatically implemented for EventType
  • For enum types, implement Event as following
use bhv::*;

enum MyEvents {
    A(i32),
    B(char),
    C,
}

impl Event for MyEvents {
    fn event_name(&self) -> &str {
        match self {
            Self::A(_) => "Event::A",
            Self::B(_) => "Event::B",
            Self::C => "Event::C",
        }
    }
}

License

Crate licensed under the MIT license.

No runtime deps

Features